Oren Data Distillation Experiment
Collection
Two identical d10 models (100M params) trained to validate the hypothesis
that quality-filtered data enables more efficient training.
•
2 items
•
Updated
•
1